Play it again, hand...
JoeMerchant writes:
USC researchers built the "Musician Hand," a four-fingered tendon-driven robot that learns to play piano by ear after just two minutes of random "motor babbling" on the keys. Hearing a ~30-note melody once, it converts the audio to a spectrogram, maps sounds to the motor commands needed to reproduce them, and plays the tune back in one attempt - well enough that blind judges sometimes couldn't tell it from trained human pianists.
The work, led by Hesam Azadjou under Francisco Valero-Cuevas at USC Viterbi (published in Royal Society Interface), challenges the traditional robotics assumption that good performance requires massive data, heavy computation, and tightly controlled environments. Instead, it mimics how animals learn: perceive, guess, adapt - using minimal energy and experience.
The researchers see the same "perceptual robotics" approach enabling cheaper, faster-deployed machines that work in unpredictable real-world settings - e.g., exoskeletons that learn an individual's gait early in Parkinson's and later help restore it, or home physical-therapy robots that adapt to each patient in real time.
Read more of this story at SoylentNews.